Object Detection in Shelf Image of the Retail store using Single Stage Detector

The primary objective is to develop a robust system for precise object identification on retail shelves, accurate item counting, and product categorization through class detection. This multifaceted approach directly tackles critical issues in retail industry, encompassing inventory management and optimized product placement. The ultimate goal is to boost operational efficiency and elevate customer satisfaction within the retail sector. The proposed methodology involves comprehensive steps, including data collection, annotation, pre-processing, and the deployment of advanced object detection models based on YOLOv5, with a particular emphasis on YOLO due to its real-time capabilities and being known as a single stage detector. By achieving these milestones, the proposed model promises to provide valuable solutions for the retail industry, reducing out-of-stock situations, minimizing manual labour, and delivering enhanced product organization. Thus, ensuring good performance compared to its other object detectors of the same use case. We have modified the architecture at different levels of the State-of-the-art YOLOv5 (Backbone, Head) and compared the models implemented for inference. We found out that Soft-IoU layer added in the head has given the best performance in detecting object with mean average precision of 79.7